18th iccrts · 2013-12-17 · this section describes techniques and/or methods that are used to...
TRANSCRIPT
18th ICCRTS
“C2 in Underdeveloped, Degraded and Denied Operational Environments”
Title of Paper: Multi-Objective Optimization for Trustworthy Tactical Networks: A Survey and Insights
Topic(s):
10: Cyberspace Management 1: Concepts, Theory, and Policy
8: Networks and Networking
Name of Author(s):
Jin-Hee Cho, Ph.D. Army Research Laboratory – CISD
Ing-Ray Chen, Ph.D. Virginia Tech – Department of Computer Science
Yating Wang Virginia Tech – Department of Computer Science
Kevin S. Chan, Ph.D. Army Research Laboratory – CISD
Ananthram Swami, Ph.D. Army Research Laboratory – CISD ananthram. [email protected]
Point of Contact:
Jin-Hee Cho, Ph.D. Army Research Laboratory – CISD
Report Documentation Page Form ApprovedOMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering andmaintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, ArlingtonVA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if itdoes not display a currently valid OMB control number.
1. REPORT DATE JUN 2013 2. REPORT TYPE
3. DATES COVERED 00-00-2013 to 00-00-2013
4. TITLE AND SUBTITLE Multi-Objective Optimization for Trustworthy Tactical Networks: ASurvey and Insights
5a. CONTRACT NUMBER
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) U.S. Army Research Laboratory,CISD,Aberdeen Proving Ground,MD,21005
8. PERFORMING ORGANIZATIONREPORT NUMBER
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES Presented at the 18th International Command & Control Research & Technology Symposium (ICCRTS)held 19-21 June, 2013 in Alexandria, VA.
14. ABSTRACT Multi-objective optimization (MOO) is the process of optimizing multiple objective functions concurrentlyand systematically. Modern Army missions require a tactical network to achieve multiple objectives asmultiple parties with different objectives are involved in collaborative mission execution. MOO problemshave been studied extensively in the field of coalition formation based on evolutionary algorithms or gametheoretic approaches. However, there has been no generic framework to consider MOO problems intactical networks particularly when the goals must be achieved based on trustworthiness of participatingentities. First, we provide a comprehensive survey of work on MOO formulation and solution techniquesparticularly in coalition formation. Second, we extensively discuss MOO techniques and methods that havebeen used for coalition formation. We specifically investigate the use of trust in the process of coalitionformation. Third, we look into tactical applications in which trust plays a pivotal role for mission success.Finally, we discuss future research directions of MOO design for trustworthy tactical networks.
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF ABSTRACT Same as
Report (SAR)
18. NUMBEROF PAGES
37
19a. NAME OFRESPONSIBLE PERSON
a. REPORT unclassified
b. ABSTRACT unclassified
c. THIS PAGE unclassified
Standard Form 298 (Rev. 8-98) Prescribed by ANSI Std Z39-18
Multi-Objective Optimization for Trustworthy Tactical
Networks: A Survey and Insights
Key words – multi-objective optimization, tactical networks, evolutionary algorithms, game theoretic approaches, trust-based applications, security, performance
Abstract – Multi-objective optimization (MOO) is the process of optimizing multiple objective functions concurrently and systematically. Modern Army missions require a tactical network to achieve multiple objectives as multiple parties with different objectives are involved in collaborative mission execution. MOO problems have been studied extensively in the field of coalition formation based on evolutionary algorithms or game theoretic approaches. However, there has been no generic framework to consider MOO problems in tactical networks particularly when the goals must be achieved based on trustworthiness of participating entities. First, we provide a comprehensive survey of work on MOO formulation and solution techniques particularly in coalition formation. Second, we extensively discuss MOO techniques and methods that have been used for coalition formation. We specifically investigate the use of trust in the process of coalition formation. Third, we look into tactical applications in which trust plays a pivotal role for mission success. Finally, we discuss future research directions of MOO design for trustworthy tactical networks.
1. Introduction
Military tactical networks are often faced with the challenges of optimizing multiple objectives where a
coalition network may have multiple partners with different objectives but under a common goal. For example, a coalition network can be formed based on cooperation or collaboration of multiple coalition
partners of different nature such as military organizations, non-government organizations (NGOs)
and/or on-site civilian organizations or entities from different nations. They may have different utilities
(or payoffs) to maximize even if they aim to collaborate for achieving a common goal. In addition, military tactical network protocols are required to be operable under resource constraints (e.g., battery life, computational power, bandwidth, and/or unreliable wireless transmission medium), lack of infrastructure (e.g., no centralized trusted entity), and/or highly hostile environments (e.g. security
vulnerability due to network or physical attacks). In order to design tactical network protocols that are
resilient against attackers, scalable under resource restrictions, and reconfigurable without any centralized trusted entities, tactical protocol designers must consider how to set and optimally achieve the multiple goals.
Multi-objective optimization (MOO) problems have been studied extensively in various multidisciplinary
domains [19] because many real-world problems in economics and engineering are usually characterized by the presence of many objectives. A common technique is to represent multiple
objectives by a single utility (or payoff) function which should be maximized. Since very often multiple objectives tend to be conflicting, the optimal solutions are not unique. As a result, MOO problems often have a set of Pareto optimal solutions referred to as the Pareto frontier. Navigation along the Pareto frontier allows designers to do trade-off analysis and select the best design for MOO [7].
In the field of coalition formation, many researchers have investigated MOO problems as one of tactical
operations based on evolutionary algorithms or game theoretic approaches. However, no generic framework exists to consider MOO problems in tactical networks particularly when the goals must be achieved based on trustworthiness of participating entities.
The concept of trust has been studied extensively in many different disciplines such as philosophy, economics, psychology, sociology, autonomic computing, and organizational management [6]. Merriam and Webster dictionary defined trust as “assured reliance on the character, ability, strength, or truth of someone or something” [31]. Cho et al. [6] has summarized the definitions of trust derived from various domains.
This work aims to comprehensively survey existing work on MOO problem formulation and solution
techniques particularly in coalition formation and to give insights on how to approach MOO issues in tactical coalition networks. In addition, we survey the use of trust in the process of coalition formation.
The rest of this paper is organized as follows. Section 2 explains the concepts of coalition formation, MOO, and MOO problems in coalition formation. Section 3 describes the key techniques/methods to solve MOO problems in coalition formation. Section 4 classifies the existing work on MOO for coalition
formation with three categories based on the types of objectives. Section 5 discusses future research
directions of MOO design for trustworthy tactical networks. Section 6 concludes the paper.
2. Multi-Objective Optimization (MOO) in Coalition Formation
2.1 Coalition Formation
According to Kahan and Rapoport [14], a coalition can be formed when three or more parties get together with a common interest that gives mutual benefits. In military tactical networks, a coalition comprises different entities such as countries and/or organizations (e.g., militaries, non-government,
non-profit, and civilian organizations) that are involved in a military operation to pursue a common goal
under a single command [30]. Many disciplines including economics, political science, mathematics, and computer science have different concepts of coalition [14]. However, the common aspect of coalition is
to benefit mutually based on trust relationships between two parties, players and a coalition party (e.g., a coalition leader).
Military tactical networks often require forming a temporary coalition in order to execute a given mission where effective and efficient asset-task assignment is critical to successful mission completion.
Under a common global objective of successful mission completion, the system may have multiple objectives to achieve while the involved parties may want to maximize their own utilities. In a typical scenario, the desirable outcome is successful mission completion while maximizing resource utilization for mission success and maintaining peaceful relationships among participating entities. This will enable the trust relationships among participating entities to be sustainable, which can positively affect future
mission execution with the same coalition members.
2.2 Multi-Objective Optimization
Most engineering applications have multiple but conflicting objectives and simultaneous optimization of all objectives may not be possible [24]. For example, in a military tactical network, a commander may want to maximize mission performance while workloads should be equally distributed over all nodes and overall resource consumption should be minimized. Multi-objective optimization (MOO) is also known as multi-criteria optimization [24].
Fig. 1 shows an example of the tradeoff observed in a MOO problem. The two conflicting goals are to maximize efficiency and to minimize resource consumption. MOO often yields a set of optimal solutions, called optimal Pareto frontiers [24].
Traditionally single objective optimization (SOO) has one objective function that may have multiple constraints. A SOO problem is often stated as follows:
(1)
The function to be optimized is where vector X indicates the set of independent input variables. Functions and describe the problem constraints [9], [11]. A MOO problem can be stated as follows:
(2)
The functions to be optimized are in , and X is the set of independent variables. Similar to the SOO problem formulation, and specify the problem constraints. Often, multiple objectives may conflict, and thus the objective solutions may be conflicting as well. A solution may optimize one objective while it may compromise other objectives [9], [11].
3. Techniques and Methods in MOO for Coalition Formation
This section describes techniques and/or methods that are used to achieve MOO in coalition formation problems. We classify MOO techniques and methods into three categories: conventional approaches, evolutionary algorithms, and game theoretic approaches.
3.1 Conventional Approaches
Conventional approaches convert a MOO problem to a SOO problem. We discuss two main techniques: weighted sum and ε-constraints.
3.1.1 Weighted Sum
This technique creates a single objective function as a linear combination of the multiple objective functions
Fig. 1: Example of Multi-Objective Optimization.
(3)
Many works use this technique for multiple criteria decision making in which each weight represents the degree or priority level of that objective function [25], [26].
3.1.2 ε-Constraints
This method constructs a single objective function where only one of the functions is optimized while the remaining functions are constraints. The objective function can be stated as [9]:
(4)
is the function selected for optimization and the other (n-1) functions are modeled as constraints [11]. Matsatsinis and Delias [20] used ε-constraints to solve task allocation problems in multi-agent
decision making systems.
3.2 Evolutionary Algorithms
Evolutionary algorithms (EAs) are categorized as metaheuristics, high-level algorithmic strategies that
direct other heuristics or algorithms while searching through the feasible solution space in order to find an optimal solution in a SOO or MOO problem [11]. This technique has been used to solve NP-complete problems such as scheduling and the traveling salesman problem. Many works have used this technique
to solve coalition formation (or task assignment or resource allocation) problems [2], [8], [9], [13], [23], [29]. This technique often finds close-to-optimal solutions in a polynomial time. Here we briefly discuss
the general structure of evolutionary algorithms.
Fig. 2: The Structure of General EAs.
Population
Individual i
Individual i+1 Aptitude Function (Fitness)
Objective Function
(SOO/MOO) Individual n
……
.
Genetic Operators
Selection
Crossover (Recombination)
Mutation
EAs have been used to solve combinatorial optimization problems whose optimal solutions can be
obtained only with a very high computational overhead. As shown in Darwin’s theory of evolution, the three fundamental components of evolution are: replication, variation, and natural selection. Replication forms a new entity from a previous one. Variations may occur during the replication. When
the fittest entities win a competition, they are selected as survivors while the weakest ones die out. Mimicking the biological evolutionary process, EAs use the generic EA structure in Fig. 2.
An individual represents a solution to the problem. A population is the set of individuals that can be used to find an optimal value. The population can be improved by genetic operators that iteratively modify individuals of the population. An aptitude function estimates the fitness value of an individual
solution. The objective function plays the role of the aptitude function. Genetic operators contribute to
improving the individuals of the population. They include selection, crossover (recombination), and mutation. Selection is the step to select the best or fittest individuals. Crossover (i.e., recombination) is the process that can combine two elements of the current population to produce one or more offspring.
Mutation transforms an individual into a new individual (i.e., a solution) [9], [11], [24].
Various types of evolutionary algorithms have been devised to solve resource assignment, such as
quantum-based evolutionary algorithms [2], hierarchical evolutionary algorithms [8], genetic algorithms
[9], [13], [23], simulated annealing [9], and hybrid particle swarm optimization algorithms [29].
3.3 Game Theoretic Approaches
Many coalition formation problems have been formulated using game theoretic approaches. We discuss two major approaches that have been popularly used: auction theory and coalition game theory.
3.3.1 Auction Theory
An auction can be performed when a seller (auctioneer) wants to sell any goods and there are buyers
(bidders) who are willing to pay the price [17]. Similarly, in the coalition formation problem, a coalition
leader wants to recruit its members to maximize its payoff and a potential bidder wants to join the coalition if the coalition gives the best gain by doing so. In this case, how to define a coalition payoff (i.e., the auctioneer’s criteria to determine winners) significantly affects the member selection process
and finally team composition. From a bidder’s perspective, it will bid an item and commit itself to buy the item based on whether buying the item gives it the best individual payoff.
Fig 3. Example of Auction Process in Hierarchical C2 Structure.
Various types of auction-based algorithms have been proposed to solve coalition formation problems in the literature such as a single-item auction with multiple preferences [5], auction-based mechanism
design for efficient bandwidth allocation [15], a reverse auction [10], and an auction with different
bidding strategies (e.g., pessimistic or optimistic) [28].
In military tactical networks where an hierarchical structure is common with a commander and several task leaders, a two-layer auction process may be modeled from a commander to task leaders and then
from task leaders to regular members. An example auction process that describes the hierarchical Command and Control (C2) structure is shown in Fig. 3. Note that decisions for bidding and winner determination (the winner selection process) are based on payoff maximization to individual bidders and task leaders.
3.3.2 Cooperative Game Theory
A cooperative game is a game in which groups of players, called coalitions, cooperate to obtain benefits
by joining a grand coalition, the set including all coalitions. The game is a competition between coalitions of players. A cooperative game is also called a “coalitional game” [22]. Often strategic games
assume that individual players are selfish and maximize their utility with no cooperation. The goal of the cooperative game is to model the situations in which the players work together or share some cost to benefit each other. However, players are selfish in that they are cooperative only if such behavior
maximizes their utility [22].
In a cooperative game, two key elements must be specified: (1) a set of players; and (2) a characteristic
function that estimates the value derived from different subsets of the players in the game. Let be the set of players and let i, where i runs from 1 to n, index each player differently. The characteristic function, denoted by , computes the value of a subset S of N, denoted by .
is the value expected when the members of S form a coalition. That is, a cooperative game can be defined as a pair where N is a finite set and is a function that maps a subset of N to a value [22].
Function can reflect the objective(s) of the system that should be attained in the coalition payoff. Each player computes its individual payoff based on its objective while a coalition estimates the expected
payoff when a subset of members is chosen. Reward (or incentive) or penalty is used to enforce an individual player’s behavior (or decision) to maximize the coalition payoff. Various versions of
cooperative games have been used to solve coalition formation problems: using repeated cooperative games [12], hedonic games [25], and nontransferable utility cooperative games [27]. It should be noted
that trust is considered in the payoff functions that impact the decisions of entities (i.e., players and coalition parties) [12].
While the MOO function is a linear combination of individual objective functions, it should be noted that
the individual objective functions are typically non-linear functions of the variables of interest. One could consider a non-linear combination of the individual objective functions, reflecting prior knowledge about the interactions of the different objectives of interest.
4. MOO Classification
We classify existing work in MOO into three classes based on the concept of global welfare (system
objectives) vs. individual welfare (individual objectives). The class 1 applies when there are no individual welfare functions. The class 2 is the case when all agents have identical individual welfare. The class 3 can be found when each agent may have its own individual welfare function.
4.1 Class 1 (C1): Global Welfare Only
MOO problems in this class deal with only multiple system/network objectives for global welfare. There
are many works dealing with multiple criteria in engineering problems [2], [8], [9], [13], [29].
Balicki [2] examined a task assignment problem in a distributed system using a quantum-based multi-
objective algorithm which adopts a new probabilistic representation called Q-bit. This work aimed to minimize workload and communication cost while maximizing system reliability. Dieber et al. [8] investigated an optimal set of configurations for a visual sensor network consisting of a large number of
camera nodes based on an evolutionary algorithm. Here the objectives are to minimize overall energy consumption and data volume while maximizing quality-of-service in terms of the camera frame rate and resolution.
Donoso and Fabregat [9] studied a team formation problem in social networks seeking the best tradeoff between skill coverage and team connectivity. Member skill coverage should meet a required skill level
while their trust relationships with other members should be close enough for active interactions. They
applied genetic algorithm and simulated annealing to identify an efficient team composition satisfying both goals. Jin et al. [13] proposed an adaptive intelligent task allocation scheme based on genetic algorithm with the goals of maximizing network lifetime based on the balance of energy consumption
among collaborative nodes while minimizing latency incurred in executing a task by providing sufficient processing power.
Yin et al. [29] examined an optimal task allocation problem in a distributed computing system where
program modules need to be allocated to different processors to maximize the system reliability under
resource constraints. This work formulated the MOO problem as a hybrid particle swam optimization problem to identify a near-optimal task allocation within a reasonable time.
4.2 Class 2 (C2) : Global Welfare and Individual Welfare
In this class of MOO problems, all agents have identical individual welfare functions. Game theoretic and market-based approaches have been proposed to solving this class of problems [3], [5], [6], [11], [16],
[17], [28].
Breban and Vassileva [3] proposed a multi-objective security game mechanism to solve a security MOO
problem. This work makes one of multiple security objectives as the main objective and uses other objectives as constraints. It investigates how Pareto frontiers may be generated for the main objective
as the other constraints vary. The system, as a defender, maintains multiple objectives based on the attacker type while an attacker takes the role of a player to maximize its payoff by achieving its goal.
Chang et al. [5] proposed a trust-based task assignment protocol that uses composite trust to select best team members to maximize the mission completion ratio while meeting an acceptable risk level where an individual entity aims to maximize its utilization. Cho et al. [6] proposed a combinatorial auction-
based solution for multiple mission assignment in MANETs where the network has two goals in terms of minimizing communication overhead caused by mission assignment and maximizing mission completion ratio while each node aims to maximize its utilization. Goel and Stander [11] proposed a trust and motivation based clan formation method where self-interested agents want to maximize their payoff by
joining, maintaining, or dissolving a clan. In this work, a coalition’s goal is to maximize its payoff by not
missing cooperative opportunities but to minimize communication cost for forming a clan.
Koloniari and Pitoura [16] formulated the problem of clustered overlay network formation as a strategic
game where nodes join a cluster based on their interest or content. The clustered overlay network has been used to efficiently exchange data relevant to the queries with less effort. An individual node has two conflicting goals: minimizing the cost for the recall of the queries with a sufficient number of cluster
memberships (as fetching information from a node in the same cluster costs less) and minimizing the cost to maintain multiple memberships. The system has multiple objectives: speed of convergence to Pareto optimality, cost optimality (minimum cost for both recall and membership), load balance
(minimum number of memberships and associated cost), and overhead minimization (movement of
nodes, social cost, etc.). Lin and Huai [18] integrated trust into market-driven resource management decision making in Grid environments to achieve resource sharing, while accurately detecting malicious entities. In this work, a consumer tries to minimize the price of resource usage for its task completion
under constraints of budget and deadline. A provider, forming a club for customers, aims to maximize the total revenue of its club and resource utilization while sharing resource only with trustworthy entities for effective resource usage. A club in this work mimics a coalition in game environments where a provider pursues global welfare.
Whitten et al. [28] presented a decentralized task assignment algorithm where each agent can make
autonomous decisions but is restricted by constraints. This work used a consensus-based bundle
algorithm for modeling different types of behavioral strategies. Both a task planner and an agent have the same goal to optimize task assignment in this work.
The commonality of MOO research in class C2 is that payoffs earned by individual entities often directly or indirectly contribute to global welfare. Therefore, in C2 MOO research, we observe that the goal of an individual entity and that of a coalition are well aligned and mutually beneficial to maximize their
payoffs.
4.3 Class 3 (C3): Global Welfare and Individual Welfare with Different Individual Payoff Functions
In this class of MOO problems, the individual payoff functions are all different. This scenario is rare because most works deal with two-party objectives such as an individual entity vs. a coalition or system. Breban and Vassileva [3] studied a long-term coalition formation problem of both vendors and
customers where each agent evaluates other agents based on trust. Each individual agent, either a vendor or a customer, joins a coalition based on its trust assessment to maximize the payoff, while the
system aims at reducing the convergence time to reaching an equilibrium state for coalition formation. This work measured trust based on positive experience in transactions and similarity in preferences. This
work is unique in that three-party objectives are being considered to solve a coalition formation problem. Meng et al. [21] presented a generic mathematical method for transforming the multi-
objective problem into a multi-player game theoretic problem.
5. Future Research Directions and Insights
In this section, we envisage an example of coalition formation for trustworthy tactical network
environments where multiple objectives may exist. A tactical network should be sustainable, that is, the system should meet the key characteristics of current/future warfare in military tactical networks. Leveraging the concept of sustainability in ecology and biology [1], a sustainable system should be
bearable (or resilient) against hostile entities, equitable towards resource utilization, and viable (or survivable) under failure or lack of resources. Reflecting those key factors, we exemplify the system goals as (1) maximizing mission completion ratio and resource utilization in the presence of hostile or faulty entities (i.e., high resilience against hostility/failure); (2) maximizing load balance among nodes in
the network (i.e., high equitability in resource usage); and (3) minimizing the delay to complete time-sensitive tasks (i.e., viability under time constraints).
We illustrate coalition formation in tactical networks with an example scenario in Fig. 4. In tactical
networks, trustworthiness of participating entities is critical to successful mission completion. A coalition consisting of qualified members tends to achieve successful mission completion when the selected members behave as expected. If members of a coalition do not behave as expected, then the mission
execution cannot continue and a new set of members should be selected for mission execution. This
process will continue until the mission deadline. If the mission cannot be completed by the deadline due
to untrustworthiness of the current members, the mission will fail. Otherwise, the mission is considered to have been executed successfully albeit with extra delay.
Fig. 4: Example Scenario of Coalition Formation in a Tactical Network.
Node trustworthiness can be evaluated from trust perspectives. Among the works surveyed in Section 4, [5], [9], [11] used trust to solve a coalition (or team, clan, alliance) formation or task assignment MOO
problem. However, except for [5], they assumed that trust is already in place and can be used as a metric to help achieve MOO. Different from the existing work, Chang et al. [5] proposes a trust-based
task assignment protocol to solve a MOO problem in a tactical coalition network. It captures trust based
on multiple dimensions of an entity’s trust under dynamically changing network environments (e.g.,
hostility, node failure or mobility), given task characteristics for tactical operations as input to the MOO problem. Trust is known to be related to risk. Hence, using trust/risk for node trustworthiness assessment for member selection and coalition formation can minimize mission abort and help satisfy
multiple system goals in mission completion ratio, minimum delay, and efficient/effective resource utilization.
We suggest some future research directions for developing coalition formation MOO algorithms for
trustworthy tactical networks as follows:
Provide a mechanistic yet repeatable method to define critical multiple objectives, given targeted tactical operations and/or mission characteristics as input;
Prioritize objectives according to mission characteristics or based on critical tradeoffs;
Develop node behavior models reflecting the behaviors of entities in the targeted network
environment;
Develop attacker models reflecting the behaviors of malicious nodes (inside attackers);
Develop trust-based MOO solution techniques allowing each entity to make decisions based on its trust assessment towards other entities, where trust assessment should take into account the unique properties of trust [6];
Model as well define payoffs (or utilities) of all involved parties that would meet multiple objectives;
Consider critical tradeoffs (e.g., minimum delay vs. fair resource sharing) in distributed and resource-constrained environments;
Devise effective reward/penalty mechanisms to entice cooperative behaviors of individual entities to increase coalition and individual payoffs; and
Devise effective and efficient optimization techniques to identify optimal solutions in resource constrained tactical network environments.
Coalition Formation
Mission Execution
Mission Completion
Mission Abort Mission Failure
6. Summary
We performed a comprehensive survey on coalition formation problems with multiple objectives and techniques and methods to solve multi-objective optimization (MOO) problems in coalition formation. We discussed three main approaches used to model and solve MOO problems in coalition formation: conventional techniques, evolutionary algorithms, and game theoretic approaches. In addition, we
classified MOO problems based on the design concept of global welfare vs. individual welfare. Trust has been used in solving coalition formation MOO problems where trust of entities may significantly affect
the performance of attaining objectives in the system. Trust has been used in solving MOO problems arising in coalition formation where trust of entities may significantly affect the attainment of system objectives. Trust can play a significant role by suggesting heuristics leading to low complexity solutions
that are not too far from the optimal solution. However, in the literature, trust has been assumed to be in place and static, which is not realistic in practice. We suggested future research directions including trust-based MOO solution techniques to
better solve coalition formation problems in tactical networks, taking into consideration the unique nature of trust and characteristics of tactical network environments.
References
[1] W. M. Adams, “The future of sustainability: rethinking environment and development in the twenty-first century,” Report of the IUCN Renowned Thinkers Meeting, Jan. 2006.
[2] J. Balicki, “An adaptive quantum-based multiobjective evolutionary algorithm for efficient task assignment in distributed systems,” Proc. World Scientific and Engineering Acadeny and Society 13th
Int'l Conf. on Computers, pp. 417-422. [3] S. Breban and J. Vassileva, “Using inter-agent trust relationships for efficient coalition formation,”
Proc. 15th Conf. of the Canadian Society for Computational Studies of Intelligence on Advances in Artificial Intelligence, pp. 221-236, Calgary, Canada, May 2002. Advances in Artificial Intelligence (Springer), Aug. 2002.
[4] M. Brown, B. An, C. Kiekintveld, F. Ordonez, M. Tambe, “Multi-objective optimization for security games,” Proc. 11th Int’l Conf. on Autonomous Agents and Multiagent Systems, vol. 2, pp. 863-870,
June 2012.
[5] M. Chang, J.H. Cho, I.R. Chen, K. Chan, and A. Swami, “Trust-based task assignment in military tactical networks,” 17th Int’l Command and Control Research and Technology Symposium, Fairfax, Virginia, June 2012.
[6] J.H. Cho, A. Swami and I.R. Chen, "A Survey on Trust Management for Mobile Ad Hoc Networks," IEEE Communications Surveys and Tutorials, vol. 13, no. 4, Nov. 2011, pp. 562-583.
[7] C. A. C. Coello, “An updated survey of GA-based multiobjective optimization techniques,” ACM Computing Surveys, vol. 32, no. 2, pp. 109-143, June 2000.
[8] B. Dieber, C. Micheloni, and B. Rinner, “Resource-aware coverage and task assignment in visual sensor networks,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 21, no. 10,
pp. 1424-1437, Oct. 2011. [9] Y. Donoso and R. Fabregat, Multi-Objective Optimization in Computer Networks Using
Metaheuristics, Auerbach Publications, Taylors and Francis Group, 2007.
[10] N. Edalat, C. Than, and W. Xiao, “An auction-based strategy for distributed task allocation in
wireless sensor networks,” Computer Communications, vol. 35, no. 8, pp. 916-928, 2012.
[11] T. Goel and N. Stander, “Multi-objective optimization using LS-OPT,” LS-DYNA Forum, Frankenthal,
Germany, 2007. [12] H. Huo, Y. Linlin, and X. Shen, “Trust game analysis between partners in the supply chain alliance,”
Int’l Conf. on Management Science and Industrial Engineering, pp. 947 – 950, 2011.
[13] Y. Jin, J. Jin, A. Gluhak, K. Moessaner and M. Palaniswami, “An intelligent task allocation scheme for multihop wireless networks,” IEEE Transactions on Parallel and Distributed Systems, vol. 23, no. 3,
pp. 444-451, March 2012.
[14] J. P. Kahan and A. Rapoport, Theories of Coalition Formation, Lawrence Erlbaum Associates, 1984. [15] M. Klein, G. A. Moreno, D. C. Parkes, D. Plakosh, S. Seuken, and K. Wallnau, “Handling
interdependent values in an auction mechanism for bandwidth allocation in tactical data
networks,”Proc. 3rd Int'l Workshop on Economics of Networked Systems (NetEcon’08), Aug. 22, 2008, Seattle, Washington, USA.
[16] G. Koloniari and E. Pitoura, “A game-theoretic approach to the formation of clustered overlay networks,” IEEE Transactions on Parallel and Distributed Systems, vol. 23, no. 4, pp. 589 - 597, April
2012. [17] V. Krishna, Auction Theory, Second Edition, Elsevier, 2010. [18] L. Lin and J. Huai, “QGrid: an adaptive trust aware resource management framework,” IEEE Systems
Journal, vol. 3, no. 1, pp. 78-90, March 2009.
[19] R. T. Marler and J. S. Arora, “Survey of multi-objective optimization methods for engineering,”
Structure Multidisc Optimization, vol. 26, no. 6, pp. 369-395, April 2004. [20] N. F. Matsatsinis and P. Delias, “AgentAllocator: An agent-based multi-criteria decision support
system for task allocation,” 1st Int'l Conf. on Applications of Holonic and Multi-Agent Systems,
Prague, Czech Republic, Lecture Notes in Computer Science, vol. 2744, pp. 1082-1083, Sept. 2003. [21] R. Meng, Y. Ye, and N. G. Xie, “Multi-objective optimization design methods based on game
theory,” 8th World Congress on Intelligent Control and Automation, pp. 2220-2227, 2010. [22] N. Nisan, T. Roughgarden, E. Targos, and V. V. Vazirani, Algorithmic Game Theory, Cambridge
University Press, Sept. 2007.
[23] C. Notario, R. Baert, and M. D’Hondt, “Multi-objective genetic algorithm for task assignment on
heterogeneous nodes,” Int’l Journal of Digital Multimedia Broadcasting, vol. 2012, article ID 716780, pp.1-12.
[24] G. P. Rangaiah, Multi-Objective Optimization: Techniques and Applications in Chemical Engineering, World Scientific Publishing, 2009.
[25] W. Saad, Z. Han, T. Basar, M. Debbah, A. Hjorungnes, “Hedonic coalition formation for distributed task allocation among wireless agents,” IEEE Transactions on Mobile Computing, vol. 10, no. 9, pp. 1327-1344, Sept. 2011.
[26] W. Saad, Z. Han, M. Debbah, A. Hjorungnes, and T. Basar, “Coalitional game theory for communication networks,” IEEE Signal Processing Magazine, vol. 26, no. 5, pp. 77-97, Sept. 2009.
[27] C. Singh, S. Sarkar, and A. Aram, “Provider-customer coalitional games,” IEEE/ACM Transactions on Networking, vol. 19, no. 5, pp. 1528 - 1542, Oct. 2011.
[28] A. K. Whitten, H. Choi, L. B. Johnson, and J. P. How, “Decentralized task allocation with coupled constraints in complex missions,” American Control Conf., pp. 1642-1649, San Francisco, CA,
June/July 2011. [29] P. Yin, S. Yu, P. Wang, and Y. Wang, “Task allocation for maximizing reliability of a distributed
system using hybrid particle swarm optimization,” Journal of Systems and Software, vol. 80, no. 5,
pp. 724-735, May 2007. [30] Coalition Networks, [Online]: http://news.bbc.co.uk/2/hi/middle_east/3873359.stm, Feb. 4, 2013.
[31] Merriam Webster’s Dictionary [Online]: http://www.merriamwebster.com/dictionary/trust%5B1%5D.
U.S. Army Research, Development and
Engineering Command
ICCRTS 2013, Paper ID: 082
Jin-Hee Cho *, Ing-Ray Chen+, Yating Wang+, Kevin Chan *, Ananthram Swami *
*US Army Research Laboratory +Virginia Tech
Multi-Objective Optimization
for Trustworthy Tactical
Networks: A Survey and
Insights
June 19-21, 2013
ICCRTS 2013
2
Outline
• Motivation
• Contributions
• Multi-Objective Optimization (MOO) in Coalition Formation
– Coalition Formation
– Multi-Objective Optimization (MOO)
• MOO Techniques for Coalition Formation
– Conventional Approaches
– Evolutionary Algorithms
– Game Theoretic Approaches
• MOO Classification based on Nature of Individual Objectives
• Current ARL Efforts
• Future Research Directions and Insights
3
Motivation
• Multiple objectives may exist in tactical networks:
– Coalition partners with different objectives
– Multiple system goals with restricted resources
• Examples of system goals are:
– Sustainability / survivability
– Resilience
– Scalability
– Reconfigurability for agility
– Resource efficiency
• Multiple goals may conflict:
– Performance vs. security
– Accuracy vs. efficiency
– Effectiveness vs. survivability
4
• Used a novel classification developed to categorize
existing work on coalition formation for MOO
• Delivered the overview of research trends in solving
coalition formation MOO problems in terms of used
techniques
• Showed the recent trends that use trust concept to solve
MOO problems in tactical networks
Contributions
5
Coalition Formation
• The common aspect of coalition is mutual benefits based on trust
relationships between two parties
• Examples:
– Asset-task assignment for successful mission completion with
multiple coalition partners
– Service composition to maximize service (mission) satisfaction in
battlefield situations
– Achieving sustainability for future performance while satisfying the
current performance based on effective/efficient resource allocation
According to Kahan and Rapoport (1984):
A coalition can be formed when three or
more parties get together with a common
interest that gives mutual benefits.
6
Multi-Objective Optimization
• An example of MOO in a military
tactical network:
– Maximize mission performance;
– Maximize load balance over all
nodes;
– Minimize overall resource
consumption
• MOO often yields a set of optimal
solutions, called optimal Pareto
frontiers
Source: http://www.enginsoft.com/
7
Single-Objective
Optimization (SOO)
• Function f(X) is to be optimized;
• Vector X indicates the set of independent input variables
• Functions H(X) and G(X) describe the problem
constraints
Multi-Objective
Optimization (MOO)
SOO vs. MOO
8
• Weighted Sum: creates a single objective function (OF) as
a linear combination of the multiple OFs
• Used in multiple criteria decision makings
• Each weight: the degree or priority level of the respective OF
• Individual OFs are typically non-linear functions of the variables
of interest
MOO Techniques for Coalition Formation
Conventional Approaches
Convert a MOO problem to a SOO problem
9 MOO Techniques for Coalition Formation
Conventional Approaches
Convert a MOO problem to a SOO problem
• ε-Constraints: constructs a single OF where only one of the
functions is optimized while the remaining functions are
constraints
• fi(X) is the function selected for optimization and the other (n-1)
functions are modeled as constraints
10 MOO Techniques for Coalition Formation
Evolutionary Algorithms
• Categorized as metaheuristics, high-level algorithmic strategies that direct other
heuristics or algorithms
• Search through the feasible solution space to find an optimal solution
• Mainly used for NP-Complete problems (e.g., combinatorial optimization prob.)
• Often finds close-to-optimal solutions in a polynomial time
11 MOO Techniques for Coalition Formation
Game Theoretic Approaches
Auction Theory
• In a coalition formation problem:
– A coalition leader wants to recruit its members to maximize its payoff;
– A potential bidder wants to join the coalition if the coalition provides
the best gain by doing so
12 MOO Techniques for Coalition Formation
Game Theoretic Approaches
Cooperative Game Theory (aka. Coalitional game)
• A cooperative game is a game in which groups of
players, called coalitions
• Player: joins a coalition that maximizes its own
individual payoff (selfish)
• Coalition leader: chooses players to maximize its own
coalition
• The goal of the cooperative game is to maximize a
grand coalition's payoff
Example of cooperative game process in hierarchical C2 structure
Compose compositions of all teams under a
mission to maximize the payoff of mission team
Select a member to maximize the payoff of a task
team by TL
Join a task team to maximize the payoff of an
individual member
13 MOO Methods and Objectives
for Coalition Formation
• Literature review for 2002-2012; 22 works
• Dominant approaches are Evolutionary algorithms and game theoretic
approaches
• Main objectives are closely related to resource constraints and system payoffs
0 1 2 3 4 5 6
evolutionary algorithm
ε-constraints
swarm optimization
auction theory
cooperative game theory
game theoretic
others
# of works
0 2 4 6 8 10
min. workload
max. reliability / survivability
min. resource consumption
max. QoS
min. risk / security vulnerability
max. payoff
others
# of works
14
MOO Classification
• Class 1 (C1): No individual objectives
• Class 2 (C2): Individuals have identical objectives
• Class 3 (C3): Individuals have different objectives
In all three classes, system objectives must also be
optimized
15
MOO Classification: C1
C1: System Objectives Only (no trust is considered)
Author(s) System/coalition objectives Techniques/Solutions Problem
Balicki (2009) Minimize workload and cost; maximize system
reliability Quantum-based evolutionary
algorithm
Task assignment
Dieber et al. (2011)
Minimize energy consumption and data volume; maximize quality-of-service Evolutionary algorithm Task allocation
Jin et al. (2012) Maximize network lifetime; minimize latency for
task execution Fitness function based on
genetic algorithms Task allocation
Matsatsinis and Delias (2003)
Maximize speediness of task execution and assignments functionality; minimize risk due to
allocation decision ε-constraints Task allocation
Notario et al. (2012)
Maximize task execution quality; minimize energy and bandwidth consumption
Genetic algorithm Task
assignment
Yin et al. (2007) Maximize reliability; Minimize resource (memory /
computational power) consumption Hybrid particle swam
optimization Task allocation
16
MOO Classification: C2
C2: System Objectives and Identical Individual Objectives (no trust is
considered)
Author(s) Individual objectives System/coalition objectives Techniques/Solutions Problem
Cho et al. (2011)
Maximize node utilization Minimize communication overhead
caused by mission assignment; Maximize mission completion
Combinatorial auction Mission
assignment
Edalat et al. (2012)
Minimize bid waiting time Minimize energy consumption and
delay in task assignment Reverse auction in cooperative game
Task allocation
Genin and Aknine (2010)
Maximize node utilization, given resource constraints
and task requirement
Maximize coalition payoff Similarity and frequency based selection
algorithms
Coalition formation
Koloniari and Pitoura
(2012)
Minimize cost for queries recall and membership
maintenance
Minimize the convergence time to optimality, load balance,
membership and recall cost, and required overhead
Cluster formation game Formation of
clustered overlays
Nardin and Sichman (2010)
Maximize a payoff as a share of the coalition payoff
Maximize throughput and valuation; minimize delay Hedonic coalition game Task allocation
Saad et al. (2011)
Customer: Maximize service satisfaction as a share of
coalition payoff
Provider: Maximize revenue in wireless network service as
coalition payoff
Nontransferable payoff coalitional game
Resource allocation
Singh et al. (2011)
An agent: maximize its utilization by being assigned
to tasks
Task planner: maximize task assignment
Consensus-based bundle algorithm
Task assignment
17
MOO Classification: C3
C3: System Objectives and Different Individual Objectives (no trust is
considered)
Author(s) Individual objectives System/coalition objectives Techniques/Solutions Problem
Meng et al. (2010)
Maximize an individual player’s
objective where each individual
has a different objective
Minimize computational cost; Maximize optimality
accuracy
Nash equilibrium, cooperative, and
evolutionary game
Generic MOO solution
18 Trust-based Solutions for
MOO Problems
Author (s) Individual Objectives System Objectives Techniques Problem
C1 Dorn et al.
(2011)
Maximize skill coverage and
team connectivity
Genetic algorithms /
simulated annealing
Team
formulation
C2 Chang et al.
(2012)
Maximize node utilization Maximize mission completion
ratio under a given a risk
constraint
Auction-based Task
assignment
Guo et al.
(2009)
Maximize membership
period
Maximize efficiency and
security in business process
Trust and self-
confidence based
Coalition
formation
Huo et al.
(2011)
Maximize an individual
payoff
Maximize the profit of the
supply chain alliance
Cooperative game Alliance
formation
Mikulski et
al.
(2011)
Maximize an individual
payoff
Maximize trust synergy;
minimize trust liability
Cooperative game Coalition
formation
Griffiths and
Luck (2003)
Maximize an individual
payoff
Maximize coalition payoff;
Minimize resource
consumption
Congregating;
cooperation-based
clan formation
Clan
formation
C3 Breban and
Vassileva
(2002)
Vendor:
Maximize sales
Customer: Minimize prices
Maximize stability of an
optimal formation of
coalitions
Trust-based coalition
formation
Coalition
formation
Trust has been used to solve coalition formation (task assignment) with
multiple objectives
19 ARL Current Effort: Trust-based
Coalition Formation for MOO
Goal: The proposed task
assignment technique is to:
• Meet multiple system objectives
using composite trust-based
member selection process
• Reduce complexity significantly for
finding close-to-optimal solutions
• Conduct comparative performance
study of the proposed technique
Composite Trust
Trust Aggregation
• Social Connectedness
• Reciprocity
• Competence
• Integrity
• Direct evidence
• Indirect evidence Comm.
networks
Social/cog.
networks
Info.
networks
Importance
Urgency
Difficulty
Task Model
Maximize mission completion ratio (PMC)
Maximize resource utilization (U)
Minimize delay to task completion (D)
Multiple Objectives PMOO = PMC + U - D
20 ARL Current Effort: Trust-based
Coalition Formation for MOO
• Higher mission completion ratio (PMC) and node utilization (U; not shown) and lower delay (D)
with more trustworthy nodes
• Higher multi-objective value (PMOO) with more trustworthy nodes
Non-trust: no trust used; ranking: trust-based; optimal: optimal solution using ILP
Cho, Chan with Chen and Wang (Virginia Tech), IEEE GLOBECOM13 submitted, March 2013
Result: Trust ranking-based selection outperforms non-trust-based scheme while performing close to the optimal solution
Composite trust based member selection improves performance with less complexity
PMOO: Multi-Objective value
PMOO = PMC + U - D
21
ARL Current Effort: Trust-based
Service Composition/Binding for MOO
Goal: The proposed service
composition and binding
technique is to:
• Meet multiple system objectives by
maximizing MOO function
• Improve performance objectives by
making trust-based decisions
• Conduct comparative performance
study of the proposed technique
and non-trust baseline scheme
Composite
Trust
Competence Integrity
Service Composition Specification (SCS)
Service selection criteria:
abstract service type Si
service requirements in terms of: Quality of Information (Q) Delay (D) Cost (C) Multiple Objectives
Maximize Quality-of-Information (Q)
Minimize Delay (D)
Minimize Cost (C)
User Satisfaction Received (USRm
): User
satisfaction level based on received
service provision over advertised service
quality
22
USTm: User satisfaction threshold for operation m
USRm: User satisfaction received based on advertised quality of service provision
Terr: trust estimation error; Pbad: % of bad nodes; Prisk: % of risk taking by malicious nodes
Cho, Chan, Swami with Chen and Wang (Virginia Tech), IEEE MILCOM13 submitted, May 2013
Trust weighted qualification assessment improves performance objectives
Result: Trust-based scheme shows higher resilience against % of malicious entities (and various intensity of malicious activities) with higher MO values / USRm
ARL Current Effort: Trust-based
Service Composition/Binding for MOO
23
Future Research Directions
• Provide a systematic yet repeatable method to
define critical multiple objectives;
• Develop node behavior (attack) models;
• Define payoffs (or utilities) of all involved parties
and/or reward/penalty mechanisms
• Devise effective and efficient MOO techniques
24
Questions?
Thank You!
U.S. Army Research Laboratory
Jin-Hee Cho
301.394.0492
2800 Powder Mill Rd.
Adelphi, MD 20783